The Reserve Bank of India (RBI) announced that it is pushing an initiative to enhance the system for handling bad loans, recommending a predicted-loss-based strategy for provisioning to be started over the next year, the Economic Times said, citing the RBI 2022-23 Annual Report.
Banks will be able to create their own credit loss models and distribute provisions over the course of five years, the news report said.
“Policy measures such as guidelines on the introduction of expected loss-based approach for provisioning are likely to be announced during 2023-24,” RBI said in its Annual Report 2022-23, cited by The Economic Times. In the discussion paper, which was published in January, RBI indicated that the estimated credit losses linked to the predicted-loss-based approach to provisioning might vary, and as a result, banks would have to organize their financial assets, such as primary loans, investments classified as “held-to-maturity” or “available for sale” or other loan commitments that are irrevocable into State 1, Stage 2 or Stage 3 categories, according to the report.
The Economic Times said the report noted that it would be up to banks to make the necessary arrangements. Classifications should be completed before the first initial recognition and on ensuing reporting dates.
According to The Economic Times report, the RBI also recommended letting the banks determine the model, but it also offered some potential issues surrounding model risk and accounting for the range of unpredictability that could result in a document.
The report added that uncertainty in financial markets in the United States and Europe would require a reevaluation of risks to the viability and resilience of the financial system amid stricter fiscal policy, the Economic Times said.
To strengthen supervisory inputs, The Economic Times said the report added that an Advanced Supervisory Analytics Group (ASAG) was developed to target cases, including social media analytics, KYC compliance and government regulation to assist with the development of machine learning models.
